The present invention relates generally to electronic memory, and more specifically to spin transfer torque (STT) magnetic tunnel junction (MTJ) storage elements having a magnetic exchange coupled composite free layer and double tunnel barriers configured to minimize the magnitude of a fast switching current (e.g., a write pulse width≤10 ns) while providing high data retention (e.g., ≥10 years).
Electronic memory can be classified as volatile or non-volatile. Volatile memory retains its stored data only when power is supplied to the memory, but non-volatile memory retains its stored data without constant power. Volatile random access memory (RAM) provides fast read/write speeds and easy re-write capability. However, when system power is switched off, any information not copied from volatile RAM to a hard drive is lost. Although non-volatile memory does not require constant power to retain its stored data, it in general has lower read/write speeds and a relatively limited lifetime in comparison to volatile memory.
Magnetoresistive random access memory (MRAM) is a non-volatile memory that combines a magnetic device with standard silicon-based microelectronics to achieve the combined attributes of non-volatility, high-speed read/write operations, high read/write endurance and data retention. The term “magnetoresistance” describes the effect whereby a change to certain magnetic states of the MTJ storage element (or “bit”) results in a change to the MTJ resistance, hence the name “Magnetoresistive” RAM. Data is stored in MRAM as magnetic states or characteristics (e.g., magnetization direction, magnetic polarity, magnetic moment, etc.) instead of electric charges. In a typical configuration, each MRAM cell includes a transistor, a MTJ device for data storage, a bit line and a word line. In general, the MTJ's electrical resistance will be high or low based on the relative magnetic states of certain MTJ layers. Data is written to the MTJ by applying certain magnetic fields or charge currents to switch the magnetic states of certain MTJ layers. Data is read by detecting the resistance of the MTJ. Using a magnetic state/characteristic for storage has two main benefits. First, unlike electric charge, magnetic state does not leak away with time, so the stored data remains even when system power is turned off. Second, switching magnetic states has no known wear-out mechanism.
STT is a phenomenon that can be leveraged in MTJ-based storage elements to assist in switching the storage element from one storage state (e.g., “0” or “1”) to another storage state (e.g., “1” or “0”). For example, STT-MRAM 100 shown in FIG. 1 uses electrons that have been spin-polarized to switch the magnetic state (i.e., the magnetization direction 110) of a free layer 108 of MTJ 102. The MTJ 102 is configured to include a reference/fixed magnetic layer 104, a thin dielectric tunnel barrier 106 and a free magnetic layer 108. The MTJ 102 has a low resistance when the magnetization direction 110 of its free layer 108 is parallel to the magnetization direction 112 of its fixed layer 104. Conversely, the MTJ 102 has a high resistance when its free layer 108 has a magnetization direction 110 that is oriented anti-parallel to the magnetization direction 112 of its fixed layer 104. STT-MRAM 100 includes the multi-layered MTJ 102 in series with the FET 120, which is gated by a word line (WL) 124. The BL 126 and a source line (SL) 128 can, depending on the design, run parallel to each other. The BL 126 is coupled to the MTJ 102, and the SL 128 is coupled to the FET 120. The MTJ 102 (which is one of multiple MTJ storage elements along the BL 126) is selected by turning on its WL 124.
The MTJ 102 can be read by activating its associated word line transistor (e.g., field effect transistor (FET) 120), which switches current from a bit line (BL) 126 through the MTJ 102. The MTJ resistance can be determined from the sensed current, which is itself based on the polarity of the magnetization direction 110 of the free layer 108. Conventionally, if the magnetization directions 112, 110 of the fixed layer 104 and the free layer 108 have the same polarities, the resistance is low and a “0” is read. If the magnetization directions 112, 110 of the fixed layer 104 and the free layer 108 have opposite polarities, the resistance is higher and a “1” is read.
When a voltage (e.g., 500 mV) is forced across the MTJ 102 from the BL 126 to the SL 128, current flows through the selected cell's MTJ 102 to write it into a particular state, which is determined by the polarity of the applied voltage (BL high vs. SL high). During the write operation, spin-polarized electrons generated in the reference layer 104 tunnel through the tunnel layer 106 and exert a torque on the free layer 108, which can switch the magnetization direction 110 of the free layer 108. Thus, the amount of current required to write to a STT-MRAM MTJ depends on how efficiently spin polarization is generated in the MTJ. Additionally, STT-MRAM designs that keep write currents small (e.g., Ic<25 micro-ampere) are important to improving STT-MRAM scalability. This is because a larger switching current would require a larger transistor (e.g., FET 120), which would inhibit the ability to scale up STT-MRAM density.
However, in order to achieve fast switching (e.g., a write pulse width 10 ns) in STT MRAM devices, a large current is needed. More specifically, in a fast switching regime, a so-called overdrive current, which is the difference between the switching current Ic at a certain pulse width and the critical current Ic0, is inversely proportionally to the write pulse, as shown in Equation (1).
                              η          ⁢                                                    I                c                            -                              I                                  c                  ⁢                                                                          ⁢                  0                                                      e                    ⁢                      t            p                          ∝                  m                      μ            B                                              Equation        ⁢                                  ⁢                  (          1          )                    
In Equation (1), Ic− Ic0 is the overdrive current, η is the spin polarization of the magnetic materials, tp is the pulse width, m is the total moment of the free layer material, and μB is the Bohr magneton, which is a constant. Equation (1) suggests that to minimize the switching current at a certain pulse width, it is necessary to reduce the free layer moment (m). Providing a low moment free layer is also advantageous for improving the MTJ's deep bit write error rate (WER) performance.
However, simple low moment free layers suffer from low activation energy (Eb), which results in poor retention. In general, the activation energy is the amount of energy required to flip the MTJ free layer's magnetic state. In order to retain data that has been written to the MTJ free layer, the activation energy must be sufficiently high to prevent random energy sources (e.g., heat) in the MTJ's operating environment from unintentionally applying enough energy to flip the MTJ free layer. Accordingly, in known STT-MRAM operating in a fast switching regime, minimizing the switching current through a low moment free layer has the undesirable result of lowering activation energy (Eb) and data retention (e.g., ≤10 years).